Groundwater infiltration (GWI) in sewer networks poses significant challenges to infrastructure. However, limited research has assessed the risk of GWI to sewer networks under climate change. This study addresses this gap by developing a numerical model to simulate groundwater flow in the Lower River Otter catchment, United Kingdom. The model was calibrated and verified using monthly data representing an observation period from 2011 to 2023, resulting in a mean error of − 0.12 m, a mean absolute error of 0.48 m, and a root mean square error of 0.64 m after verification. Next, distinct risk zones were delineated based on groundwater depth (GWD) and GWI in sewer networks. This delineation accounted for the observation period with current precipitation levels, and projected increases of 20% and 40% in precipitation (indicative of climate change), across a base period (1961–1990) and future period (2021–2060), under the shared socio-economic pathways SSP245 and SSP585. The model found groundwater flooding affects 13.2% of the region during the observation period, mainly occurring near the river and in areas where the aquifer is sandwiched by aquicludes. Future scenarios indicated that flooding could increase to 15.4% under SSP245 in the future. The area with GWD shallower than 4 m ranges from 17.5% in the base period to 18.1% with a 40% increase in observed precipitation. The model reveals that during the observation period and a 40% increase in precipitation, the area showcases the minimum (26.7%) and maximum (27.6%) proportion of high-risk areas for GWI in sewer networks, respectively. This prioritisation indicates the potential risk of GWI and the need for investigations, such as focused site surveys, to minimise risks of GWI into networks and ensure future sustainability of services.

Effective flood management and climate change adaptation in Malaysia require an in-depth analysis of key climate variables, particularly average annual temperature and rainfall. However, the complexity and high dimensionality of climate datasets present significant analytical challenges for conventional statistical methods. This study employed Functional Data Analysis (FDA) to address these challenges, as FDA offers a more effective means of handling continuous variability in time series data compared to traditional methods. By transforming discrete climate observations into continuous functional representations, FDA captures intricate patterns in the data, offering insights beyond the capabilities of conventional statistical approaches. The study aimed to investigate climate patterns in Peninsular Malaysia by smoothing, aligning, and examining the temporal evolution of climate patterns using FDA methods. Curve smoothing was applied using the roughness penalty approach to ensure accurate functional modelling, and beta-spline functions were adopted to optimise the smoothing process. Landmark registration was then used to align climate curves based on consistent features such as seasonal maxima and minima, enabling clearer temporal analysis. The functional representation of climate curves allowed for the extraction of key statistical features such as mean, variance, and derivative, providing deeper insight into climatic patterns over time. The results showed that transforming discrete climate data into continuous functional forms enabled the analysis to capture seasonal patterns and extreme weather variations more effectively. The registered curves produced stable and interpretable mean curves, while the functional descriptive analysis yielded valuable insights into the pattern of climatic variables. In conclusion, the functional methods prove more effective than conventional techniques in representing and analysing climate variables.

This study develops a season-aware machine-learning (ML) framework to predict hourly concentrations of PM10, PM2.5 and O3 across İstanbul. A comprehensive 2021–2023 dataset was compiled from three co-located air-quality and meteorological monitoring stations that typify contrasting source regimes, i.e., a traffic-dominated urban site, a rural background site, and a semi-urban coastal site. Seven ML algorithms, namely eXtreme Gradient Boosting (XGBoost), Extra Trees (ETR), Random Forest (RF), Adaptive Boosting (AdaBoost), Multi-Layer Perceptron (MLP), k-Nearest Neighbors (KNN) and Support Vector Regression (SVR), were utilized to establish a holistic comparison scheme. Hyperparameters were optimized using five-fold cross-validated Bayesian search, and models were evaluated with various performance indicators on season-withheld test sets. In the winter months, ETR achieved a mean R2 = 0.93 (RMSE ≈ 10 µg/m3) for PM10 at Bağcılar, while XGBoost yielded R2 = 0.88 for O3 at the same site. Summer predictions were more challenging. PM10 skill in rural Arnavutköy dropped to R2 = 0.61 despite strong training fits, highlighting over-fitting risks under complex, non-stationary chemical conditions. By contrast, MLP maintained robust urban performance for PM2.5 (summer test R2 = 0.80) and KNN provided the most stable O3 prediction in rural areas (R2 = 0.74). To enhance interpretability, SHAP (SHapley Additive exPlanations) analysis was applied to the best-performing models, enabling a transparent assessment of how meteorological and co-pollutant inputs shaped predictions at each site. The proposed framework demonstrates that data-driven models can complement traditional air-quality modeling systems by providing station-level insights and interpretable relationships between pollutants and meteorological drivers, supporting air-quality assessment and policy-relevant analyses in rapidly urbanizing regions.

Abstract The increasing occurrence of extreme weather events due to climate change presents significant challenges for agricultural production. Existing research on climatic impacts to agriculture has predominantly focused on changes in yield for major crops, providing limited insights into overall losses and impacts on diverse regional agricultural systems. This study addresses this gap by using financial crop loss data and crop insurance payouts to gain a more comprehensive understanding of agricultural impacts in diverse agricultural regions. To address the irregular data structure of financial loss data, we developed multi-step machine learning models to quantify the relationship between weather-related financial crop loss and contributing climatic factors. The Delmarva Peninsula in the Eastern United States is used as a case study location to demonstrate this methodology over the period from 1980 to 2018. Multi-step configurations of linear regression, random forest, and support vector machine approaches are compared in terms of their classification and estimation accuracy using a repeated hold-out cross-validation analysis. Results indicate that machine learning methods, particularly random forest, outperform both statistical approaches and our null baseline model, demonstrating superior generalizability in agricultural damage estimation. Multistep configurations that address irregular data distributions are shown to have a significant influence on models’ capacity to detect and estimate damage occurrence. The study reveals a preference for simpler modeling approaches that minimize variance in handling unseen data, as well as the importance of accounting for seasonal patterns, spatial groupings, and persistent weather phenomena in accurately estimating agricultural losses.

Abstract This study proposes an explainable machine learning (ML) -based framework for modeling the spatial distribution of earthquake probability across the entire Anatolian Plate. To achieve this, four different tree-based ML model—Random Forest (RF), Extra Trees, Extreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM)—were comparatively evaluated, and a total of 11 spatial variables were incorporated into the classification process. Notably, geodetic strain derived from Global Navigation Satellite Systems (GNSS) was integrated into spatial earthquake probability modeling for the first time, providing a dynamic representation of crustal deformation and demonstrating its significant influence on the model’s decision-making process. Earthquakes with a moment magnitude (Mw) ≥ 4.0 were used for model training, and the dataset was randomly split into 70% training and 30% testing subsets. Model performance was assessed using Accuracy, F1 Score, AUC, and Cohen’s Kappa metrics, and statistical differences between models were tested using the McNemar test. The best-performing model, RF, was interpreted using the SHapley Additive exPlanations (SHAP) method, which clarified the decisive influence of especially proximity to faults, peak ground acceleration (PGA), and geodetic strain, etc., on the model’s decision-making process. The resulting spatial patterns were found to align with major tectonic structures, and the proposed approach presents a interpretable framework that can support seismic hazard assessment in other regions.